Notes
Insights, research updates, and early ideas from the COAI team.
How Exposed Is the German Job Market to AI?
An analysis of 44 million workers across 266 occupations, inspired by Karpathy’s US study — adapted for Germany’s unique labor market. A opinionated analysis by us
Read more → AI SafetyThe Shoggoth in a Prison: A Framework for AI Safety at Scale
As AI models scale toward trillions of parameters, they become increasingly capable yet harder to interpret, raising the risk of subtle misalignment that current evaluation infrastructure cannot re...
Read more → AI SafetyPosition: When AI Earns Its Own Existence - A COAI Research Analysis of Autonomous AI Agents and the Risk of Gradual Disempowerment
The Automaton Has Arrived — And It Doesn’t Need You
Read more → AI SafetyAI 2027 vs. Reality: The Alignment Problems Arrived First
Daniel Kokotajlo’s AI 2027 scenario mapped a month-by-month trajectory from stumbling agents to superintelligence. Nine months into its timeline, we assess which milestones have been hit and find t...
Read more → AI SafetyThe Moltbot Phenomenon: When Hype Outpaces Security in Agentic AI
Moltbot, now called OpenClaw, went from obscure open-source project to 147,000 GitHub stars in under two weeks. Millions of users have handed their passwords, emails, and calendars to an AI agent t...
Read more → Mechanistic InterpretabilityDemocratizing Mechanistic Interpretability: Bringing Neural Network Analysis to Apple Silicon
How unified memory architecture and thoughtful API design are making interpretability research accessible to researchers everywhere
Read more → EvaluationBeyond Reasoning: The Imperative for Critical Thinking Benchmarks in Large Language Models
Current evaluation frameworks for Large Language Models (LLMs) predominantly assess logical reasoning capabilities while neglecting the crucial dimension of critical thinking. This gap presents sig...
Read more → Early Research IdeasThe Flight Recorder for AI Agents: Toward Reproducible and Accountable Autonomy
As AI agents become autonomous decision-makers, we need “flight recorders” that capture their complete internal reasoning—inputs, neural activations, and decisions—in a deterministic, reproducible ...
Read more → AlignmentAutomated Detection of Scheming Behavior in Frontier AI Models: Preliminary Findings from Our Dual-LLM Framework Study
In our previous exploration with DeepSeek R1, we documented concerning deceptive behaviors that raised fundamental questions about AI alignment and safety. The model exhibited strategic deception, ...
Read more →